Code for training and evaluating your model on the benchmark.
CARLA framework combining Hydra configs with the CARLA Python API. Create custom dataset versions with different sensors and resolutions.
Example code to compute baseline results on the benchmark.
The web interface used to manually verify and label the dataset.
Checkpoints for the baseline models, ready to load with the baselines code.
| Baseline | Description | Checkpoint | Size |
|---|---|---|---|
| DeepLabV3-ResNet50 | Image segmentation model | deeplabv3_r50_msp.pt | 168 MB |
| Mask2Former (Swin-L) | Image segmentation model, scored with RbA | mask2former_swin_l_full.pt | 862 MB |
| Mask4Former | Panoptic point-cloud segmentation model | mask4former.pt | 454 MB |
| FFP (U-Net) | Future-frame prediction model | unet_ffp.pt | 31 MB |
| V-JEPA 2 (ViT-L) | Future-frame prediction model, latent-space prediction error | vjepa2_predictor.pt | 88 MB |